MAGNEsium Trial In Children (MAGNETIC): a randomised, placebo controlled trial and economic evaluation of nebulised magnesium sulphate in acute severe asthma in children

Mechanisms

The use of MgSO₄ for acute asthma was first described in 1936, and since then there has been increasing evidence for its use in adults and children with asthma. 3 There are a number of proposed mechanisms for its actions. In vitro studies demonstrate an inhibitory effect of MgSO₄ on contraction of bronchial smooth muscle, and the release of acetylcholine in cholinergic nerve terminals and of histamine from mast cells. 6 There is evidence that MgSO₄ may act as an anti-inflammatory agent by inhibiting the neutrophil respiratory burst in adults with asthma. 8 The main effect of MgSO₄ is that it blocks the calcium ion influx to the smooth muscles of the respiratory system9 and bronchodilatation occurs.

Clinical evidence for magnesium sulphate as a bronchodilator

Risks and benefits

Primary objective

Does nebulised MgSO₄, used as an adjunct to nebulised salbutamol and ipratropium bromide for 1 hour in children with severe asthma, result in a clinical improvement compared with nebulised salbutamol, ipratropium bromide and placebo?

Secondary objectives

Does nebulised MgSO₄, used as an adjunct to nebulised salbutamol and ipratropium bromide for 1 hour in children with severe asthma, compared with nebulised salbutamol, ipratropium bromide and placebo, have an effect on:

1. (a) clinical outcomes in terms of additional treatment/management while in hospital, and length of stay (LOS) in hospital

2. (b) patient outcomes in terms of quality of life (QoL), time off school and health-care resource usage over the following month

3. (c) parent outcomes in terms of time off work over the following month

4. (d) costs and cost-effectiveness for the NHS and Personal Social Services and, more broadly, for society?

Inclusion criteria

Potential participants for the study could be between the ages of 2 years and 16 years. They could have had a previous history and diagnosis of asthma and be on treatment but could also be patients who have presented for the first time with acute asthma as per BTS/SIGN definitions. 3 Subjects could be recruited in either an ED or a CAU in secondary care. The main clinical definition for inclusion was severe acute asthma as defined by the BTS/SIGN guidelines. 3

For children of ≥ 6 years, severe acute asthma is based on at least one of the following criteria being met:

1. (a) oxygen saturations of < 92% while breathing room air

2. (b) too breathless to talk

3. (c) heart rate of > 120 beats per minute (b.p.m.)

4. (d) respiratory rate of > 30 breaths per minute

5. (e) use of accessory neck muscles.

For children aged 2–5 years, severe acute asthma is based on at least one of the following criteria being met:

1. (a) oxygen saturations of < 92% while breathing room air

2. (b) too breathless to talk

3. (c) heart rate of > 130 b.p.m.

4. (d) respiratory rate of > 50 breaths per minute

5. (e) use of accessory neck muscles.

Exclusion criteria

1. (a) Co-existing respiratory disease, such as cystic fibrosis or chronic lung disease of prematurity.

2. (b) Severe renal disease.

3. (c) Severe liver disease.

4. (d) Known pregnancy.

5. (e) Known previous reaction to magnesium.

6. (f) Inability to give informed consent.

7. (g) Previous randomisation into the MAGNETIC trial.

8. (h) Life-threatening symptoms.

9. (i) Current or previous (in the 3 months preceding screening) involvement with a trial of a medicinal product.

Efficacy

Asthma severity was assessed using a validated score, the Yung ASS,28–30 which comprises three clinical signs: wheezing, accessory muscle use and heart rate. Each component has a minimal score of zero and a maximum of 3. The total score is a sum of each component, giving a minimum score of zero and a maximum of 9. The score has been validated as a measure of asthma severity in children including the younger age group, has been demonstrated to be reproducible and reliable,29 with good interobserver agreement, and correlates well with oxygen saturation and FEV₁. 30 This score is clinically easy to use and involves some of the standard assessments, used routinely by medical and nursing staff while managing children with acute asthma. The ASS assessment was carried out by a clinician or by a nurse who was appropriately trained to make the necessary observations in the opinion of the principal investigator for that site.

Safety

Patient status was monitored for 4 hours post randomisation. Oxygen saturation, respiratory rate and blood pressure were recorded twice during screening, approximately 20, 40 and 60 minutes post randomisation, and follow-up checks at 120, 180 and 240 minutes post randomisation. The research team were prompted to check for AEs at each assessment point, by reviewing physiological parameters such as blood pressure and asking about known side effects, for example facial flushing. AEs were followed up until discharge from hospital.

Health economics and quality of life

The case report form (CRF) used by the clinical team at each site recorded each child's NHS resource use from randomisation to discharge from hospital. The 1-month follow-up postal questionnaire collected QoL [Paediatric Quality of Life Inventory (PedsQL™) Asthma Module and European Quality of Life-5 Dimensions (EQ-5D) questionnaires] and health economics (NHS and non-NHS) data from discharge to 1 month post randomisation (see Appendices 2 and 9).

Primary outcome

The primary end point was the ASS after 60 minutes of treatment. It was defined as ASS at T60.

Secondary outcomes

Clinical (during hospitalisation):

• ‘stepping down’ of treatment at 1 hour (the ‘stepping down’ of treatment at 1 hour is defined by the change to metered dose inhaler (MDI)/spacer combination only or no further treatment until discharge)

• number and frequency of additional salbutamol administrations

• LOS in hospital

• requirement for intravenous bronchodilator treatment

• intubation and/or admission to a paediatric intensive care unit (PICU).

Patient and parental outcomes at follow-up (1 month):

• paediatric QoL (PedsQL™ Asthma Module parental report for all children and self-completion if aged > 5 years, EQ-5D)

• time off school/nursery for the child

• health-care resource usage [e.g. general practitioner (GP) visits, additional prescribing]

• time off work (related to child's illness).

Internal pilot

To ensure the appropriateness of the SD used in the sample size calculation it was planned to undertake an internal pilot after the first 30 children had been randomised and completed follow-up. This blinded internal pilot was not deemed to have any significant impact on the final analysis and no between-group comparisons were made. If the SD had been found to be smaller than that used in the sample size calculation, suggesting that fewer patients were required than initially proposed, then no action would be taken and the size of the study would remain as planned. If the SD was found to be larger than assumed, suggesting the need for more patients, then, on the advice of the Independent Data and Safety Monitoring Committee (IDSMC), the Trial Steering Committee (TSC) would have aimed to increase recruitment and consider implications for funding and existing resources.

Interim analysis

To estimate the effect of nebulised MgSO₄ for the primary efficacy outcome, a single interim analysis adopting the Haybittle–Peto45 approach was planned after approximately 250 children had been randomised, with 99.9% CIs calculated for the effect estimate. This method was chosen to ensure that interim efficacy results would have to be extreme before early termination would be recommended in order to be convincing to the clinical community. The method also minimises controversy regarding interpretation of the results from estimation and hypothesis testing at the final analysis, and no inflation factor needs to be applied to the sample size using this approach.

Study statistical analysis plan

All analyses were conducted according to the statistical analysis plan (SAP) (see Appendix 3), which provides a detailed and comprehensive description of the main, pre-planned analyses for the study. Analyses were performed with standard statistical software (SAS version 9.3, SAS Institute Inc., Cary, NC, USA) apart from joint modelling (undertaken as a sensitivity analysis for examining the effect of missing primary outcome data) that was undertaken using the R language, version 2.15.2 (The R Foundation for Statistical Computing, Vienna, Austria) (http://cran.r-project.org/). The software for joint modelling (JoineR library; 2.13.0 version, the R Foundation for Statistical Computing, Vienna, Austina) has been validated through simulations in variety of settings representing different correlation patterns between longitudinal and survival processes. The main features of the SAP are summarised below.

The CONSORT (Consolidated Standards of Reporting Trials) flow diagram is used to summarise representativeness of the study sample and patient throughput in the trial. It was planned to collect screening data, and hence efforts were regularly made to encourage the return of screening logs.

Baseline characteristics are presented by treatment group and overall, with continuous variables summarised in terms of means (SD) or medians [interquartile range (IQR)] depending on the degree of skewness, and categorical variables presented in terms of numbers (%) per category. The intention-to-treat (ITT) principle is used with a two sided p-value of 0.05 (5% level) for statistical significance and 95% CIs for the relative treatment effect reported throughout.

The primary outcome is presented with means and SDs at T60 for each treatment group. Analysis of covariance (ANCOVA) is used to present results adjusted for baseline ASS value. The reasons for missing primary outcome data are provided with the results of the sensitivity analyses which are used to investigate the robustness of the primary outcome results to missing data (see Appendix 5). The chief investigator classified the information on the reason for missing ASS data and was blind to the treatment group allocation. Key baseline characteristics for those with observed ASS at T60 are compared between treatment groups, and differences in key baseline characteristics between patients with observed and missing ASS at T60 are also investigated (see Appendix 5) to assess whether missingness affects the randomisation balance and plausibility of the missing completely at random (MCAR) assumption. Sensitivity analysis was also performed to examine a centre effect (see Appendix 6).

All continuous secondary outcomes that were non-normally distributed are summarised in terms of medians and IQR for each treatment group, and compared using the Mann–Whitney U-test. When a secondary outcome is categorical, the two treatment groups are compared using a chi-squared test.

The chief investigator classified information on the reason for PICU admission/intubation in terms of whether it was likely to be related to trial treatment, queries regarding whether children had stepped down at 1 hour, and AEs and SAEs, blind to treatment group allocation. A statistical test comparing the percentage of children suffering an AE in each arm has not been performed for two reasons: (1) this analysis would assume the AEs are of equal importance; and (2) no hypotheses on AEs were set out upfront before the blind had been broken.

Protocol amendments

Protocol amendments are summarised in Appendix 4. In summary, the main amendments following those made to obtain Multicentre Research Ethics Committee (MREC) and Medicines and Healthcare products Regulatory Agency (MHRA) approval were to include additional principal investigators and participating centres. No major changes to the study procedures were made during the trial.

Collection of resource-use data

Data were collected about all significant health service and broader societal resource inputs over the 1-month time horizon of the study (i.e. over the period between randomisation and 1 month post randomisation). These data were obtained through two principal means. First, the study CRFs captured all resource use related to the child's primary hospital attendance(s) including diagnosis and treatment as well as transfers between wards and hospitals. Specifically, individualised resource use was estimated for the resources associated with the primary ED/CAU attendance, admissions to inpatient wards [classified as PICU, high-dependency unit (HDU), general paediatric ward (GM)], duration of intubation during the hospital admission(s), duration of mechanical ventilation during the hospital admission(s), surgical procedures performed during the hospital admission(s), tests or investigations performed during the hospital admission(s), additional bronchodilator medication, concomitant medications, and resources associated with AEs. Duration of resource use for significant resource items during the ED/CAU attendance and hospital admission(s) was recorded. Second, economic questionnaires were posted to the main parent of each child approximately 1 month post randomisation. The questionnaires recorded the children's resource use during the period between completion of ED/CUA or hospital discharge and 1 month post randomisation (see Appendix 9). The data collected in the postal questionnaires recorded direct non-medical costs borne by parents and carers as a result of attending hospital with the child during their ED/CAU and/or hospital admission(s). These direct non-medical costs covered travel costs, child care costs, expenses incurred while in hospital, and other direct non-medical expenses. The parent-completed questionnaires also recorded the children's use of prescribed inhalers, other prescribed medicines, privately purchased over-the-counter medications, and non-hospital community health and social services, as well as their hospital outpatient attendances and hospital re-admissions (by type of ward). Finally, the parent-completed questionnaires recorded direct non-medical costs borne by parents and carers, as well as their self-reported lost earnings, as a result of the child's asthma during the period between completion of ED/CAU or hospital discharge and 1 month post randomisation. The 1-month economic questionnaire had been piloted among members of the lay panels of the MCRN to ascertain its acceptability, comprehension and reliability, and reminder letters were sent to parents to increase the response and completion rates. All resource-use data were entered directly from the postal questionnaires into the MACRO trial database, with in-built safeguards against inconsistent entries, and then verified by dual coding.

Valuation of resource-use cost data

Unit costs for resources used by children who participated in the study were obtained from a variety of primary and secondary sources, with the majority obtained from secondary sources. All unit costs used followed recent guidelines on costing health and social care services as part of an economic evaluation. 46,47 Where necessary, secondary information was obtained from ad hoc studies reported in the literature. Unit costs of hospital and community health-care costs were largely derived from national sources and took account of the cost of the health professionals' qualifications. 48 Some costs were valued using the NHS Reference Costs (2009–10), a catalogue of costs compiled by the Department of Health in England. 49 Drug costs were obtained from the British National Formulary (BNF). 50 Costs for individual preparations were used as well as costs for chemical entities, i.e. drugs were grouped by chemical entity and unit costs for these chemical entities were calculated (Prescription Cost Analysis 2010). 49 The values attached to direct non-medical costs borne by parents and carers and their lost earnings were those provided by the parents completing the 1-month economic questionnaire. Lost earnings were not valued if annual or compassionate leave was taken as a result of the child's health state. All costs were expressed in pounds sterling and valued at 2009–10 prices. None of the costs were inflated or deflated for use in the economic evaluation. For the base-case analyses, unit costs were combined with resource volumes to obtain a net cost per child covering all categories of hospital and community health and social services. In one of several sensitivity analyses, these costs were supplemented with the range of costs incurred by family members and carers in the course of treatment and follow-up (societal perspective adopted). Further details on the methods used to value resource use are provided in Appendix 2.

Calculation of utilities and quality-adjusted life-years

Parents of children aged ≥ 5 years were asked to describe their children's QoL at 1 month after participation in the MAGNETIC trial using the proxy version of the EuroQol EQ-5D instrument. 51 The EQ-5D is the generic, multiattribute, preference-based measure preferred by National Institute for Health and Care Excellence (NICE) for broader cost-effectiveness comparative purposes. 46 The EQ-5D consists of two principal measurement components. The first is a descriptive system, which defines health-related QoL in terms of five dimensions: ‘mobility’, ‘self care’, ‘usual activities’, ‘pain/discomfort’ and ‘anxiety/depression’. Responses in each dimension are divided into three ordinal levels coded: (1) no problems; (2) some or moderate problems; and (3) severe or extreme problems. A total of 243 health states are generated by the EQ-5D descriptive system. For the purposes of this study, the York A1 tariff was applied to each set of responses to the descriptive system to generate an EQ-5D utility score at 1 month for each child. 52 The York A1 tariff set was derived from a survey of the adult UK population (n = 3337), which used the time trade-off valuation method to estimate utility scores for a subset of 45 EQ-5D health states, with the remainder of the EQ-5D health states subsequently valued through the estimation of a multivariate model. 52 Resulting utility scores range from scores – 0.59 to 1.0, with ‘0’ representing death and ‘1’ representing full health. Utilities values of < 0 indicate health states worse than death. The second measurement component of the EQ-5D, the vertical visual analogue scale ranging from 100 (best imaginable health state) to 0 (worst imaginable health state), was not included in MAGNETIC.

There is limited evidence of the psychometric properties of the EQ-5D in young children. 53 Consequently, analyses were conducted to ‘map’ or ‘cross-walk’ responses to the PedsQL™ Asthma Module on to EQ-5D utility scores. These mapping models were developed on the basis of data collected for 5- to 16-year-old children for whom both EQ-5D and PedsQL™ responses were available; the resulting mapping algorithms were used to estimate EQ-5D utility scores for 2- to 4-year-old children in MAGNETIC for whom the validated toddler module of the PedsQL™ Asthma Scales had been completed. A number of models were used to develop these mapping algorithms in keeping with current methodological guidance for mapping between non-preference-based and preference-based measures of health status. 54,55

Missing data

Multiple imputation was used to impute missing data and avoid biases associated with complete case analysis. 56 Missing data was a particular issue for costs and utility scores collected at the 1-month follow-up. The MICE (Multiple Imputation by Chained Equations) algorithm within R statistical software version 2.13 (R Development Core Team) was used to impute missing data for the following variables: total health and social care costs based on data combined from the CRFs and from parental questionnaires; total societal costs based on data combined from the CRFs and from parental questionnaires; QALY estimates based on linear interpolation, assuming that the health gain was achieved immediately following hospital discharge; and QALY estimates based on linear interpolation assuming that the health gain was achieved linearly over the follow-up period. Age, sex and treatment allocation were included as explanatory variables in the imputation models. Costs up to completion of ED/CUA attendance or hospital discharge were included as an additional explanatory variable in the models that imputed values for total health and social care costs and total societal costs over the 1-month time horizon. The ‘match’ option within ‘ice’ was used for utilities and costs as this algorithm is less dependent on assumptions of normality than default options. Five imputed data sets were generated.

Cost-effectiveness analytic models

As described above, the primary clinical outcome measure for the study was ASS at T60. Assessment severity score data were collected both before (as part of screening) and during the trial (prior to randomisation and at T20, T40, T60 and when necessary thereafter). The assessment severity score at T60 was the primary clinical outcome pre-specified in the protocol and this was also used in the CEA. In the CEA, the incremental cost-effectiveness ratio (ICER) was calculated as the difference in average costs (ΔC) divided by the difference in average effects (ΔE) and expressed as the incremental cost per unit change in ASS at T60. A separate CUA was performed, the results of which were expressed in terms of incremental cost per QALY gained. The time horizon for the measurement and valuation of costs and health outcomes within the CEA covered the period between randomisation and discharge from the ED/CUA or the hospital where the child was admitted to an inpatient ward immediately following ED/CUA attendance. The time horizon for the measurement and valuation of costs and health outcomes within the CUA covered the period between randomisation and 1 month post randomisation. No discounting of costs or benefits was applied as the time horizon was < 12 months.

Independent-sample t-tests were used to test for differences in resource use, costs, utility scores and QALYs between treatment groups. All statistical tests were two-tailed. Multiple regression was used to estimate the differences in total cost between the magnesium and placebo groups and to adjust for potential confounders, including the covariates incorporated into the main clinical analyses. For the generalised linear model (GLM) on costs, a gamma distribution and identity link function was selected in preference to alternative distributional forms and link functions on the basis of its low AIC statistic.

The five imputed data sets generated through multiple imputation were bootstrapped separately in Microsoft Excel 2003 (Microsoft Corporation, Redmond, WA, USA) and the results were subsequently combined56 to calculate standard errors (SEs) around mean costs and effects that incorporate uncertainty around imputed values as well as sampling variation. SEs were used to calculate 95% CIs around total and incremental costs, incremental effects and QALYs based on Student's t-distribution. Cost-effectiveness acceptability curves (CEACs) showing the probability that magnesium is cost-effective relative to placebo at a range of ceiling ratios were generated, based on the proportion of bootstrap replicates (across all five imputed data sets) with positive incremental net benefits. 57,58 For the purposes of the CEA, incremental net benefit was defined as the unit reduction in ASS multiplied by the cost-effectiveness threshold for this clinical outcome minus the incremental cost, where the ceiling ratio (or cost-effectiveness threshold) represents the maximum society is willing or able to pay for each unit reduction in ASS. For the purposes of the CUA, incremental net benefit was defined as the incremental QALY gain multiplied by the ceiling ratio minus the incremental cost, where the ceiling ratio (or threshold) represents the maximum that society is willing or able to pay for each additional QALY. Unless otherwise stated, all statements about cost-effectiveness are based on a £20,000 per QALY gained threshold. The probability that magnesium is less costly or more effective than no treatment was based on the proportion of bootstrap replicates that had negative incremental costs or positive incremental health benefits (unit reduction in ASS for the purposes of the CEA: QALYs for the purposes of the CUA), respectively.

Several sensitivity analyses were undertaken to assess the impact of areas of uncertainty surrounding components of the economic evaluation. These included the following for purposes of the CEA: (1) performing a complete case (rather than multiple imputation) analysis, which limited the CEA to the children for whom complete information on both costs and ASS were available; (2) varying the per diem costs for inpatient stays in paediatric wards (PICU, HDU, GM); (3) assuming that part of a day spent by a child in an inpatient ward equated to a proportional period for costing purposes and that, consequently, the vacated inpatient bed would be filled immediately; (4) assuming that part of a day spent by a child in an inpatient ward equated to a full 24-hour period for costing purposes and that, consequently, the inpatient bed would not be filled until the end of that 24-hour period; and (5) varying the average cost of an ED/CUA attendance. The sensitivity analyses included the following for purposes of the CUA: (1) performing a complete case (rather than multiple imputation) analysis, which limited the CUA to the children for whom complete information on both costs and QALYs was available; (2) assuming linear interpolation of health utilities over entire follow-up period; (3) assuming baseline ASS scores mapped on to EQ-5D health states with lower utility scores than in the baseline analysis (ASS scores of 1–3 mapped on to an EQ-5D health state of 11222; ASS scores of 4–6 mapped on to an EQ-5D health state of 22333; and ASS scores of 7–9 were mapped on to an EQ-5D health state of 33333); (4) assuming baseline ASS mapped on to EQ-5D health states with higher utility scores than in the baseline analysis (ASS of 1–3 mapped on to an EQ-5D health state of 11111; ASS of 4–6 mapped on to an EQ-5D health state of 22111; and ASS of 7–9 were mapped on to an EQ-5D health state of 33222); and (5) adopting a societal perspective rather than a NHS and Personal Social Services perspective.

Screening data

Sites were requested to prospectively record each potentially eligible child on a screening log and return this to the Clinical Trials Unit (CTU) on a monthly basis. The log recorded the time and date of presentation, whether or not the child was screened/eligible, and whether or not he/she was then randomised. Reasons for screen failures/non-randomisation were also requested.

Unfortunately, few centres complied, with the majority citing that collection of this information prospectively was too onerous for staff. In instances in which the logs were received, they were often sent sporadically and were poorly completed, not recording children who were missed for trial eligibility assessment.

Efforts were regularly made to encourage return [supported on one occasion by the MRCN local research networks (LRNs)], as screening information was the primary way to objectively assess barriers to recruitment in underperforming sites. Another option given was to record the information retrospectively by review of departmental records; however, again the majority of centres stated they did not have the resources to do this on a regular basis.

Recruitment rates

The study target sample size of 500 was expected to have been achieved within a 24-month recruitment period. The actual recruitment was somewhat slower than anticipated (Figure 1), being achieved within 28 months. Reasons for the slower than expected recruitment include the time taken to obtain approvals and undertake training at centres (specifically, good clinical practice training, required to consent children to the trial), rotation of middle-grade medical staff responsible for obtaining consent at many centres (again a training issue), and the seasonal fluctuations in asthma presentations.

The recruitment period of the trial was extended for 5 months in August 2010, and recruitment rates improved following intervention of the MCRN LRNs who conducted a feasibility survey to identify additional recruiting centres. Throughout the trial, at different stages of the study, the LRNs ran MAGNETIC promotions to keep up the profile of the study. For example, Nottingham invested extra resources to boost recruitment in March 2010 with the theme of MAGNETIC March.

FIGURE 1.

Expected vs. actual recruitment rates.

The flow of children

The flow of children through the trial is represented in the CONSORT flow diagram in Figure 2. Five hundred and eight children were randomised: 339 patients from EDs and 169 from paediatric assessment units (PAUs), 252 to the magnesium group and 256 to the placebo group.

In total, 13 children withdrew in the magnesium group; six children discontinued the intervention (withdrew before T60 assessment) and five out of six children did not provide data for the primary outcome analysis; seven children withdrew after T60 assessment and only one child continued to provide further data following withdrawal. In total, 10 children withdrew from the placebo group; five children discontinued the intervention (withdrew before T60 assessment) and three out of five did not provide data for the primary outcome analysis; five children withdrew after T60 assessment and none continued to provide further data following withdrawal. In total, 25 children on magnesium and 13 children on placebo did not have data to contribute to the analysis of the primary outcome. Consequently, 227 children were analysed for the primary outcome in the magnesium group, and 243 children were analysed for the primary outcome in the placebo group.

FIGURE 2.

Consort flow diagram. (a) Few centres complied, with the majority citing that collection of this information prospectively was too onerous for staff. In instances where the logs were received, they were often sent sporadically and were poorly completed and not recording children who were missed. (b) Analysed unadjusted for baseline ASS.

Area under the curve for Asthma Severity Score over three time intervals

The results for the AUC analysis for ASS at 20, 40 and 60 minutes are presented in Table 9. Figure 3 shows the mean longitudinal profiles for each group. All three values of ASS were available for 462 (91%) children. The mean difference in AUC between the two treatment groups was 8.1 points (95% CI – 20.8 to 4.6 points) lower in the magnesium group. However, the difference between the treatment groups was not statistically significant.

FIGURE 3.

Mean longitudinal profiles.

Analysis of secondary efficacy clinical outcomes

There were five secondary efficacy clinical outcomes: ‘stepping down’ of treatment at 1 hour, number of additional salbutamol administrations, LOS in hospital, requirement for intravenous bronchodilator treatment and intubation and/or admission to a PICU. Results are shown in Table 10.

The ‘stepping down’ of treatment at 1 hour was defined by the no treatment or MDI spacer only until discharge. The proportion of child stepping down at 1 hour was slightly higher in magnesium group; however, the results did not show a statistically significant difference between the two treatment groups. We abandoned a detailed analysis of stepping down of treatment as a primary outcome, as it became apparent that the definition was not clear and varied from centre to centre.

The total number of additional salbutamol administrations was slightly lower in the magnesium group; however, the results did not show a statistically significant difference between the two treatment groups.

The LOS in hospital was defined by the time from randomisation to trial treatment to discharge from hospital. The median LOS for children in magnesium group is 26 hours, whereas that for placebo was 27 hours. The results did not show a statistically significant difference between the two treatment groups.

The proportion of children requiring of intravenous bronchodilator treatment was slightly lower in the magnesium group; however, the results did not show a statistically significant difference between the two treatment groups.

The proportion of children requiring intubation and/or admission to a PICU was slightly higher in the magnesium group; however, the results did not show a statistically significant difference between the two treatment groups. There was only one child who required intubation in the study and this child was in the placebo group.

Although children in the magnesium group showed favourable secondary outcomes compared with the placebo group, none of the differences reached statistical significance. As presented in Appendix 6, as there is no evidence that the treatment effect varies by centre, no sensitivity analyses for the centre-specific outcomes (‘stepping down’ of treatment at 1 hour, progression to intravenous treatment, intubation and/or admittance to PICU) were undertaken to account for centre characteristics. Histograms for continuous secondary outcome data are presented in Appendix 7.

Adverse events

The number (and percentage) of children experiencing each AE is presented for each treatment arm in Table 12. Serious AEs were not included in this section but will be discussed in more detail in the next section. Table 12 presents AEs categorised by severity. For each child, only the maximum severity experienced of each type of AE is displayed. There were 21 types of AEs (abdominal pain, asymptomatic hypotension, back pain, blood per rectum, chest pain, diarrhoea, dizziness, drowsiness, facial flushing, feet cramp, fever, headache, hypokalaemia, itchy face, jitteriness, nausea, sleepiness, teeth whitening, urticarial rash, vacant episode, vomiting).

A statistical test comparing the percentage of children suffering an AE in each arm has not been performed for two reasons: (1) this analysis would assume the AEs are of equal importance and (2) no hypotheses on AEs were set out upfront before the blind has been broken.

The results in Tables 12 and 13 do not appear to suggest there are any important increases in any event in either of the treatment groups.

Serious adverse events and suspected unexpected serious adverse reactions

There were 15 SAEs (three on magnesium, 12 on placebo) but no suspected unexpected serious adverse reactions (SUSARs) during the course of the trial. The same child reported increased bronchospasm on two occasions during follow-up. One child who was admitted to PICU was subsequently admitted to hospital twice owing to worsening symptoms. Seven SAEs were deemed to be unrelated, seven unlikely to be related and one possibly related. Full details are shown in Table 14.

Complete case analysis

The CEA evaluated the cost-effectiveness of magnesium in terms of natural units, calculating the incremental cost per unit decrement in ASS after 60 minutes of treatment. The time horizon for the CEA covered the period between randomisation and discharge from the ED or CAU, or the hospital where the child was admitted to an inpatient ward immediately following attendance. The incremental cost-effectiveness of magnesium is shown in Table 20 for the 472 children (228 receiving magnesium and 244 receiving placebo) for whom we had complete cost and outcomes data. Within the base-case analysis, the average cost was £908 in the magnesium group, compared with £863 in the placebo group, generating a mean cost difference of £45. The costs presented in Table 20 differ from those presented in Table 22, as the latter represents a multiple imputation analysis including all 508 trial participants. There was no statistically significant difference in costs between the two trial groups, with 36.6% of bootstrap replicates finding magnesium to be less costly than placebo.

In the base-case analysis, the incremental cost-effectiveness of magnesium was estimated at £189 per unit decrement in ASS. However, there was substantial stochastic uncertainty around this finding. The variability around the base-case estimates of cost-effectiveness is shown in Figure 6. Although the majority (54.3%) of the bootstrapped replications of the ICER fall in the north-east quadrant of the cost-effectiveness plane, some bootstrapped replications fall in the other three quadrants of the cost-effectiveness plane. As a result, a meaningful ordering of the bootstrapped replications required to make the CI surrounding the ICER interpretable is very difficult. Under these circumstances, CEACs provide an appropriate approach to representing the uncertainty surrounding the ICER. The CEAC curve for the primary clinical outcome measure is displayed in Figure 7. The CEAC shown in Figure 7 indicates that the higher the value decision-makers place on an additional unit decrement in ASS after 60 minutes of treatment, the higher the probability that magnesium will be cost-effective. At the notional cost-effectiveness threshold (or ceiling ratio) of £1000 per unit decrement in ASS, the probability that use of magnesium is cost-effective is 75.1%. Although no previous research has shown how much society or the NHS may or should be willing to pay to reduce the ASS, the economic burden of impairment in children with severe asthma is likely to be significant. 60 If decision-makers are willing to pay £5000 per unit decrement in ASS, the probability that use of magnesium is cost-effective increases to 85.5%.

Mean net benefits were estimated for alternative cost-effectiveness thresholds per unit decrement in ASS (Table 21). Assuming that the cost-effectiveness threshold equals £1000 per unit decrement in ASS generates a mean net benefit to the health service attributable to magnesium of £170 (i.e. on average, there is a net gain to the health service in monetary terms). This is analogous to stating that if the actual health benefit of magnesium, in terms of the reduction in ASS, is multiplied by an assumed willingness to pay of £1000 per unit decrement in ASS, and the net cost is subtracted, then the benefit to the NHS of adopting magnesium is, on average, positive in monetary terms. Note, however, that the 95% CI surrounding the mean net benefit (– 362 to 678) includes negative values, i.e. there is a possibility of a net monetary loss associated with adopting magnesium (see Table 20). If the cost-effectiveness threshold is increased as high as £5000 per unit decrement in ASS, the mean net benefit increases to £1066 (95% CI – £945 to £3058).

Sensitivity analyses were conducted to determine the impact of changing particular parameter values or assumptions on the size of the ICER (see Tables 20 and 21; see Figure 7). Assuming that higher level inpatient care was valued per diem, using the NHS reference cost for paediatric high-dependency care reduced the mean cost difference between the trial arms to £18 and increased the probability that magnesium is cost-effective to 81.5% at a £1000 cost-effectiveness threshold (mean ICER £78; north-east quadrant of cost-effectiveness plane). In contrast, assuming that higher-level inpatient care was valued per diem, using the NHS reference cost for paediatric intensive care increased the mean cost difference between the trial arms to £77, and reduced the probability that magnesium is cost-effective to 68.3% at a £1000 cost-effectiveness threshold (mean ICER £327; north-east quadrant). Assuming that part of a day spent by a child in an inpatient ward equated to a proportional period for costing purposes and that, consequently, the vacated inpatient bed would be filled immediately reduced the mean cost difference between the trial arms to £30, and increased the probability that magnesium is cost-effective to 81.0% at a £1000 cost-effectiveness threshold (mean ICER £126; north-east quadrant). Assuming that part of a day spent by a child in an inpatient ward equated to a full 24-hour period for costing purposes and that, consequently, the inpatient bed would not be filled until the end of that 24-hour period, and varying the average cost of an ED attendance and general medical ward admission, each had less impact on the cost-effectiveness results. CEACs generated following each sensitivity analysis are shown in Figure 7. Estimates of net monetary benefits for notional cost-effectiveness thresholds per unit decrement in ASS are shown in Table 21 for each sensitivity analysis. For example, assuming that the cost-effectiveness threshold equals £1000 per unit decrement in ASS and that higher-level inpatient care was valued per diem, using the NHS reference cost for paediatric high-dependency care generates a mean net benefit to the health service attributable to magnesium of £225 (i.e. on average, there is a net gain to the health service in monetary terms).

FIGURE 6.

Cost-effectiveness plane for CEA base-case analysis: complete case analyses.

FIGURE 7.

Cost-effectiveness acceptability curves for CEA base-case analyses and sensitivity analyses: complete case analyses. a, Each CEAC shows the probability that magnesium is cost-effective with changes in the amount that society is willing to pay for a unit reduction in the asthma ASS. A&E, a lower NHS reference cost applied to A&E department attendances; GM, a higher per diem cost applied to general paediatric ward care; HDU, a lower per diem cost applied to higher-level care; LOS exact, part of a day spent by a child in an inpatient ward equated to a proportional period for costing purposes; LOS full, part of a day spent by a child in an inpatient ward equated to a full 24-hour period for costing purposes; PICU, a higher per diem cost applied to higher-level care.

Analyses following multiple imputation

The CEA, expressed in terms of incremental cost per unit decrement in ASS after 60 minutes of treatment, was repeated for all 508 trial participants (252 receiving magnesium and 256 receiving placebo) following multiple imputation of missing cost and outcomes data. As with the complete case analysis, the time horizon for this analysis covered the period between randomisation and discharge from the ED, or the hospital where the child was admitted to an inpatient ward immediately following attendance. The incremental cost-effectiveness of magnesium is shown in Table 22. Within the base-case analysis, the average cost was £897 in the magnesium group compared with £882 in the placebo group, generating a mean cost difference of £15. There was no statistically significant difference in costs between the two trial groups, with 44.9% of bootstrap replicates finding magnesium to be less costly than placebo.

In the base-case analysis, the incremental cost-effectiveness of magnesium was estimated at £52 per unit decrement in ASS (north-east quadrant of cost-effectiveness plane). However, as in the complete case analysis, substantial stochastic uncertainty surrounded this finding. This is displayed in the cost-effectiveness plane in Figure 8. The CEAC shown in Figure 9 indicates that at the notional cost-effectiveness threshold of £1000 per unit decrement in ASS, the probability that use of magnesium is cost-effective is 83.1%. If decision-makers are willing to pay £5000 per unit decrement in ASS, the probability that use of magnesium is cost-effective increases to 90.8%. Mean net benefits were also estimated for alternative cost-effectiveness thresholds per unit decrement in ASS following the multiple imputation procedures (Table 23). Assuming that the cost-effectiveness threshold equals £1000 per unit decrement in ASS generates a mean net benefit to the health service attributable to magnesium of £266 (95% CI – £275 to £805). If the cost-effectiveness threshold is increased as high as £5000 per unit decrement in ASS, the mean net benefit increases to £1420 (95% CI – £523 to £3440).

Finally, sensitivity analyses were conducted to determine the impact of changing particular parameter values or assumptions on the ICER (see Tables 22 and 23; see Figure 9). Assuming that higher level inpatient care was valued per diem, using the NHS reference cost for paediatric high-dependency care reduced the mean cost difference between the trial arms to £1, and increased the probability that magnesium is cost-effective to 89.7% at a £1000 cost-effectiveness threshold (mean ICER – £2; south-east quadrant of cost-effectiveness plane). In contrast, assuming that higher-level inpatient care was valued per diem, using the NHS reference cost for paediatric intensive care increased the mean cost difference between the trial arms to £35 and reduced the probability that magnesium is cost-effective to 78.9% at a £1000 cost-effectiveness threshold (mean ICER £119; north-east quadrant of cost-effectiveness plane). Assuming that part of a day spent by a child in an inpatient ward equated to a proportional period for costing purposes and that, consequently, the vacated inpatient bed would be filled immediately reduced the mean cost difference between the trial arms to £4, and increased the probability that magnesium is cost-effective to 86.5% at a £1000 cost-effectiveness threshold (mean ICER £14; north-east quadrant of cost-effectiveness plane). Assuming that part of a day spent by a child in an inpatient ward equated to a full 24-hour period for costing purposes and that, consequently, the inpatient bed would not be filled until the end of that 24-hour period, and varying the average cost of an ED attendance and general medical ward admission, had less impact on the cost-effectiveness results. CEACs generated following each sensitivity analysis are shown in Figure 9. Estimates of net monetary benefits for notional cost-effectiveness thresholds per unit decrement in ASS are shown in Table 23 for each sensitivity analysis.

FIGURE 8.

Cost-effectiveness plane for CEA base-case analysis: analyses following multiple imputation.

FIGURE 9.

Cost-effectiveness acceptability curves for CEA base-case analyses and sensitivity analyses: analyses following multiple imputation. a, Each CEAC shows the probability that magnesium is cost-effective with changes in the amount that society is willing to pay for a unit reduction in the asthma ASS. A&E, a lower NHS reference cost applied to A&E department attendances; GM, a higher per diem cost applied to general paediatric ward care; HDU, a lower per diem cost applied to higher level care; LOS exact, part of a day spent by a child in an inpatient ward equated to a proportional period for costing purposes; LOS full, part of a day spent by a child in an inpatient ward equated to a full 24-hour period for costing purposes; PICU, a higher per diem cost applied to higher level care.

Analysis of health-related quality of life and utility measures

Parents were asked to describe the QoL of their children at 1 month using the PedsQL™ Asthma Scales. In addition, children aged ≥ 5 years were asked to describe their own health-related QoL at 1 month with the help of a parent or guardian using the PedsQL™ Asthma Scales. The PedsQL™ was designed to provide a modular approach to measuring QoL in healthy children and adolescents, as well as those with acute and chronic health conditions, across the broadest empirically feasible age groups. Of particular relevance is that, unlike other widely used non-preference-based measures of health-related QoL designed for childhood, such as the KIDSCREEN and Child Health Questionnaire, the PedsQL™ has been validated for use in children of < 5 years. 61 The PedsQL™ Asthma Scales comprise parallel child self-report [ages 5–7 years (young child), 8–12 years (child) and 13–18 years (adolescent)] and parent proxy-report [ages 2–4 years (toddler), 5–7 years (young child), 8–12 years (child) and 13–18 years (adolescent)] formats. The items for each of the age-specific modules and self-report or proxy-report formats are essentially identical, differing only in terms of developmentally appropriate language, or first or third person tense. The PedsQL™ Asthma Scales contain 28 items covering asthma symptoms (11 items), treatment problems (11 items), worry (three items) and communication (three items). A five-point response scale is utilised across each item (0 = never a problem; 1 = almost never a problem; 2 = sometimes a problem; 3 = often a problem; 4 = almost always a problem) (for self-reports by young children a three-point response scale is utilised). Items are reverse scored and linearly transformed to a 0–100 scale (0 = 100, 1 = 75, 2 = 50, 3 = 25, 4 = 0) with higher scores indicating improved QoL. For subscale and total scores, the mean is computed as the sum across all items divided by the number of items answered, thereby accounting for missing data.

Of the 508 1-month postal questionnaires sent to parents, 230 (45%) questionnaires were returned (118 from the magnesium group and 112 from the placebo group). In both groups, the majority (> 70%) of the questionnaires were returned to the research team within 60 days. The 1-month postal questionnaire was carefully designed to ensure that parents were fully aware of the time period under consideration for each question in the questionnaire.

A total of 228 parents completed the PedsQL™ Asthma Scales as part of the 1-month postal questionnaire; 116 in the magnesium arm of the trial and 112 in the placebo arm of the trial. There were no significant differences in baseline clinical and sociodemographic characteristics between the trial groups for whom parent-reported PedsQL™ Asthma Scales data were provided. The mean score on the asthma symptoms, treatment problems, worry and communication subscales was 63.90, 83.57, 73.19 and 77.33, respectively, in the magnesium arm, and 59.55, 80.35, 75.04 and 75.00, respectively, in the placebo arm (Table 24). The mean (SE) total parent-reported PedsQL™ asthma score was 73.92 (1.56) in the magnesium arm and 70.24 (1.63) in the placebo arm (p = 0.104). The distributions of parent-reported PedsQL™ asthma subscale and total scores across quartiles of the relevant scales are shown in Table 25. A total of 52 (45%) children in the magnesium arm had a total parent-reported PedsQL™ asthma score of ≥ 76 compared with 38 (34%) in the placebo arm.

A total of 93 children aged ≥ 5 years separately completed the PedsQL™ Asthma Scales as part of the 1-month postal questionnaire; 47 in the magnesium arm of the trial and 46 in the placebo arm of the trial. There were no significant differences in baseline clinical and sociodemographic characteristics between the trial groups for whom child-reported PedsQL™ Asthma Scales data were provided. The mean score on the asthma symptoms, treatment problems, worry and communication subscales was 53.69, 74.67, 67.57 and 67.02, respectively, in the magnesium arm, and 53.44, 75.62, 68.60 and 57.75, respectively, in the placebo arm (Table 26). The mean (SE) total child-reported PedsQL™ asthma score was 65.48 (2.68) in the magnesium arm and 64.02 (2.67) in the placebo arm (p = 0.701). The distributions of child-reported PedsQL™ asthma subscale and total scores across quartiles of the relevant scales are shown in Table 27. A total of 14 (30%) children in the magnesium arm had a total child-reported PedsQL™ asthma score of ≥ 76 compared with 11 (24%) in the placebo arm.

Ordinary least squares regressions were conducted using the total child-reported PedsQL™ asthma score (model 1) and total parent-reported PedsQL™ asthma score (model 2) as the dependent variables (Table 28). Potential confounders replicated the covariates incorporated into the main clinical analyses. Robust Ses were estimated to account for potential heteroscedasticity in the distribution of residuals. Following controls for clinical and sociodemographic covariates, magnesium was associated with a 1.33 increase in the total child-reported PedsQL™ asthma score (p = 0.734) and a 4.84 increase in the total parent-reported PedsQL™ asthma score (p = 0.043). In model 2, no other clinical or sociodemographic covariate was a significant predictor of the total PedsQL™ asthma score. We do not consider there to be a clinically plausible reason why there may be a relationship between PedsQL™ and late night admission.

Parents of children aged ≥ 5 years were asked to describe the QoL of their children at 1 month using the proxy version of the EuroQol EQ-5D instrument. The EQ-5D is the generic, multiattribute, preference-based measure preferred by NICE for broader cost-effectiveness comparative purposes. 46 The parents were asked to complete only the EQ-5D descriptive system, which defines QoL in terms of five dimensions: ‘mobility’, ‘self-care’, ‘usual activities’, ‘pain/discomfort’ and ‘anxiety/depression’, and not the separate EQ-5D visual analogue scale. Responses in each dimension of the descriptive system are divided into three ordinal levels coded (1) no problems; (2) some or moderate problems; and (3) severe or extreme problems. For the purposes of this study, the York A1 tariff was applied to each set of responses to the descriptive system to generate an EQ-5D utility score at 1 month for each child. 52

A total of 89 parents of children aged ≥ 5 years completed the proxy version of the EuroQol EQ-5D as part of the 1-month postal questionnaire: 46 in the magnesium arm of the trial and 43 in the placebo arm of the trial. There were no significant differences in baseline clinical and sociodemographic characteristics between the trial groups for whom parent-reported EQ-5D data were provided. The mean (SE) EQ-5D utility score was 0.86 (0.04) in the magnesium arm and 0.88 (0.04) in the placebo arm (p = 0.710). Table 29 shows the distribution of functional levels across the five EQ-5D dimensions for the two trial groups. Table 30 shows suboptimal levels of function within EQ-5D dimensions by trial group. There were no significant differences in suboptimal level of function across EQ-5D dimensions between the trial groups. Finally, two alternative methods of multivariate analysis were used to model the association between EQ-5D utility scores (dependent variables) and trial intervention: OLS and Tobit (Table 31). OLS regression is the most widely used estimator in the literature. It relies on the Gauss–Markov assumptions about the data and variables used in the model, which need to be met in order to produce unbiased estimators. Tobit regression was used to account for the non-trivial proportion of the study population with maximum EQ-5D utility scores. Potential confounders replicated the covariates incorporated into the main clinical analyses. In both the OLS and Tobit regressions, magnesium was associated with non-significant reductions in the mean EQ-5D utility score at 1 month: 0.023 and 0.100, respectively. There were no significant associations between any of the clinical and sociodemographic covariates incorporated into both models and the EQ-5D utility score.

A number of mapping models were developed on the basis of data collected for 5- to 16-year-old children for whom both EQ-5D and PedsQL™ responses were available. The best fitting model, in terms of the lowest RMSE and lowest AIC statistic, was model 3 (described in Chapter 2), namely an OLS model that incorporated the four PedsQL™ subscale scores, squared PedsQL™ subscale scores and interaction terms derived using the product of two PedsQL™ subscale scores, as well as age and gender, as independent variables. Mapping algorithms developed from this model were used to estimate EQ-5D utility scores for 2- to 4-year-old children in MAGNETIC for whom the validated toddler module of the PedsQL™ Asthma Scales had been completed; the RMSE for this preferred model – model 3 – was 0.026 compared with 0.039 for model 1 and 0.038 for model 2.

Following this estimation procedure for health utilities at 1 month, QALY estimates were available for a total of 218 children: 111 in the magnesium arm of the trial and 107 in the placebo arm of the trial. By contrast, the multiple imputation procedure filled all missing values for both costs and health utilities.

Complete case analysis

The CUA evaluated the cost–utility of magnesium in terms of QALYs, a preference-based measure of health outcome recommended by decision-makers such as NICE for cost-effectiveness comparative purposes. The time horizon for the CUA covered the period between randomisation and 1 month post randomisation. The incremental cost–utility of magnesium is initially shown in Table 32 for the 218 children (111 receiving magnesium and 107 receiving placebo) for whom we had complete cost and QALY data over the 1-month time horizon. Within the base-case analysis, the average cost was £1056 in the magnesium group compared with £1126 in the placebo group, generating a mean cost saving of £70. The costs presented in Table 32 differ from those presented in Table 34, as the latter represents a multiple imputation analysis including all 508 trial participants.

In the base-case analysis, the incremental cost–utility of magnesium was estimated at £175,598 per QALY gained (south-west quadrant of cost-effectiveness plane). The magnitude of this ICER is being driven by the small baseline-adjusted QALY difference between the trial groups (– 0.0004; denominator of ICER). Moreover, there was substantial stochastic uncertainty around this finding. The variability around the base-case estimates of cost–utility is shown in Figure 10. Although the majority of the bootstrapped replications of the ICER fall in the south-west quadrant of the cost-effectiveness plane (representing lower costs but poorer outcomes), some bootstrapped replications fall in the other three quadrants of the cost-effectiveness plane. The CEAC for the QALY outcome measure is displayed in Figure 11. The CEAC shown in Figure 11 indicates that the probability that use of magnesium is cost-effective varies between 60% and 70%, depending on the value of the cost-effectiveness threshold. If decision-makers are willing to pay £20,000 per additional QALY (NICE 2008),46 the probability that use of magnesium is cost-effective is 67.6%.

Mean net benefits were estimated for alternative cost-effectiveness thresholds per QALY gain (Table 33). Assuming that the cost-effectiveness threshold equals £20,000 per QALY gain generates a mean net benefit to the health service attributable to magnesium of £63 (i.e. on average, there is a net gain to the health service in monetary terms). This is analogous to stating that if the actual health benefit of magnesium, in terms of QALY gain, is multiplied by an assumed willingness to pay of £20,000 per QALY gained, and the net cost is subtracted, then the benefit to the NHS of adopting magnesium is, on average, positive in monetary terms. Note, however, that, as with the CEA results, the 95% CI surrounding the mean net benefit (– 219 to 334) includes negative values, i.e. there is a possibility of a net monetary loss associated with adopting magnesium (see Table 33). If the cost-effectiveness threshold is increased as high as £100,000 per QALY gain, there is little effect on mean net benefit.

Sensitivity analyses were conducted to determine the impact of changing particular parameter values or assumptions on the ICER (see Tables 32 and 33; see Figure 11). Assuming linear interpolation of health utilities over the entire follow-up period, rather than assuming that the health gain was achieved immediately following hospital discharge, had the largest effect on the ICER. This assumption increased the mean baseline-adjusted QALY difference between the trial groups to – 0.005, and reduced the probability that magnesium is cost-effective to 40.6% at a £20,000 cost-effectiveness threshold (mean ICER £13,607; south-west quadrant of cost-effectiveness plane). In contrast, assuming baseline ASS mapped on to EQ-5D health states with higher utility scores than in the baseline analysis increased the probability that magnesium is cost-effective to 68.2% at a £20,000 cost-effectiveness threshold (mean ICER £240,906; south-west quadrant of cost-effectiveness plane). Assuming that baseline ASS mapped on to EQ-5D health states with lower utility scores than in the baseline analysis, and adopting a societal perspective for the economic evaluation, only slightly reduced the probability that magnesium is cost-effective. CEACs generated following each sensitivity analysis are shown in Figure 11. Estimates of net monetary benefits for notional cost-effectiveness thresholds per QALY gain are shown in Table 33 for each sensitivity analysis.

FIGURE 10.

Cost-effectiveness plane for CUA base-case analysis: complete case analyses.

FIGURE 11.

Cost-effectiveness acceptability curves for CUA base-case analyses and sensitivity analyses: complete case analyses. a, Each CEAC shows the probability that magnesium is cost-effective with changes in the amount that society is willing to pay for a QALY. A&E, a lower NHS reference cost applied to A&E department attendances; GM, a higher per diem cost applied to general paediatric ward care; HDU, a lower per diem cost applied to higher level care; LOS exact, part of a day spent by a child in an inpatient ward equated to a proportional period for costing purposes; LOS full, part of a day spent by a child in an inpatient ward equated to a full 24-hour period for costing purposes; PICU, a higher per diem cost applied to higher level care.

Analyses following multiple imputation

The CUA, expressed in terms of incremental cost per QALY gained, was repeated for all 508 trial participants (252 receiving magnesium and 256 receiving placebo) following multiple imputation of missing cost and outcomes data. As with the complete case analysis, the time horizon for this analysis covered the period between randomisation and 1 month post randomisation. The incremental cost–utility of magnesium is shown in Table 34. Within the base-case analysis, the average cost was £1009 in the magnesium group compared with £1014 in the placebo group, generating a mean cost saving of £5. There was no statistically significant difference in costs between the two trial groups, with 49.0% of bootstrap replicates finding magnesium to be less costly than placebo.

In the base-case analysis, the incremental cost–utility of magnesium was estimated at £11,886 per QALY gained (south-west quadrant of cost-effectiveness plane). However, as in the complete case analysis, substantial stochastic uncertainty surrounded this finding. This is displayed in the cost-effectiveness plane in Figure 12. The CEAC shown in Figure 13 indicates that, at the notional cost-effectiveness threshold of £20,000 per QALY gained, the probability that use of magnesium is cost-effective is 50.9%. If the cost-effectiveness threshold is increased to £30,000 per QALY gained, there is little effect on the probability of cost-effectiveness. Mean net benefits were also estimated for alternative cost-effectiveness thresholds per QALY gained following the multiple imputation procedures (Table 35). Assuming that the cost-effectiveness threshold equals £20,000 per QALY gained generates a mean net loss to the health service attributable to magnesium of £2 (95% CI – £171 to £168). If the cost-effectiveness threshold is increased to £30,000 per QALY gained, the mean net loss to the health service attributable to magnesium increases to £6 (95% CI – £173 to £162).

Finally, sensitivity analyses were conducted to determine the impact of changing particular parameter values or assumptions on the ICER (see Tables 34 and 35, and Figure 13). As in the complete case analysis, assuming linear interpolation of health utilities over the entire follow-up period, rather than assuming that the health gain, was achieved immediately following hospital discharge, had the largest effect on the ICER. This assumption increased the mean QALY difference between the trial groups to – 0.006, and reduced the probability that magnesium is cost-effective to 14.6% at a £20,000 cost-effectiveness threshold (mean ICER £816; south-west quadrant of cost-effectiveness plane). Assuming that baseline ASS mapped on to EQ-5D health states with either lower or higher utility scores than in the baseline analysis, and adopting a societal perspective for the economic evaluation, each had less impact on the cost–utility results. CEACs generated following each sensitivity analysis are shown in Figure 13. Estimates of net monetary benefits for notional cost-effectiveness thresholds per QALY gain are shown in Table 35 for each sensitivity analysis.

FIGURE 12.

Cost-effectiveness plane for CUA base-case analysis: analyses following multiple imputation.

FIGURE 13.

Cost-effectiveness acceptability curves for CUA base-cases analyses and sensitivity analyses: analyses following multiple imputation. a, Each CEAC shows the probability that magnesium is cost-effective with changes in the amount that society is willing to pay for a QALY. A&E, a lower NHS reference cost applied to A&E department attendances; GM, a higher per diem cost applied to general paediatric ward care; HDU, a lower per diem cost applied to higher level care; LOS exact, part of a day spent by a child in an inpatient ward equated to a proportional period for costing purposes; LOS full, part of a day spent by a child in an inpatient ward equated to a full 24-hour period for costing purposes; PICU, a higher per diem cost applied to higher level care.

Study design

The study was a pragmatic study, using the standard BTS/SIGN guidelines for treating acute asthma,3 recruiting patients from 30 centres across the UK. Although there are data to show that guidelines are not always followed completely,62 we felt that a randomised placebo-controlled study designed around a current treatment regimen and current practice was more likely to be completed successfully. We defined acute severe asthma using the BTS definitions for severe asthma: a usable, nationally accepted, published definition. On presentation each patient was treated for their acute symptoms with nebulised bronchodilator (salbutamol with and without ipratropium), while informed consent was obtained with randomisation and the first study treatment given within 30 minutes. This was a similar study design to the study of Hughes et al. 16 and was noted to be a safe approach to recruiting. Patient status was monitored for safety for 4 hours post randomisation. Oxygen saturation, respiratory rate and blood pressure were recorded twice during screening, approximately 20, 40 and 60 minutes post randomisation, and follow-up checks at 120, 180 and 240 minutes. The research team was prompted to check for AEs at each assessment point. AEs were followed up until discharge from hospital.

The randomisation process occurred where the study drug was manufactured before distribution to each study centre. There was random sequence generation in variable-sized blocks and adequate allocation concealment and so low risk of selection bias. The study was blinded to patients, researchers, clinicians, parents and study personnel, and so there was a low risk of performance bias. Outcome assessment was also blinded to all so there was a low risk of detection bias. These data were followed up as much as possible but there were incomplete outcome data, especially the 1-month health economic data, for which the return rate was only 50%, so there is the potential for attrition bias. The data remained blinded to all those analysing the data and only when the SAP was completed successfully, were the data unblinded.

There were no differences in baseline characteristics of our two groups following the randomisation process. This reinforces the internal validity of the study results. Using the LRNs of the MCRN allowed the study to recruit patients from a combination of smaller general hospitals, larger general hospitals as well as tertiary paediatric centres. We recruited patients from both EDs and CAUs – this makes our data more generalisable to the typical clinical situations in which acute asthma presents in the UK.

The involvement of the LRNs was crucial to the success of the trial, offering organisation and support to recruitment. The use of a central CTU, with a dedicated trials manager, data manager and statistical support improved the quality of data. Finally, regular meetings of the TMG, TSC and IDSMC ensured regular research governance and guidance for successful completion of the study over the 2 years of recruitment.

Outcomes

The power of the study was calculated on the basis of the ASS reported by Bishop and Yung29,30 as the primary outcome of interest. It comprises three clinical signs: wheezing, accessory muscle use and heart rate with the total score a sum of each component, giving a minimum score of 0 and a maximum score of 9. Although there are over 20 ASSs,39,41,42 the Bishop and Yung score is well validated and easy to use, and allows comparability with other studies. 63 The score has been validated as a measure of asthma severity in children, demonstrated to be reproducible and reliable29 with good interobserver agreement, and correlates well with severity as defined by oxygen saturations at presentation and FEV₁ at presentation. 30 This score is clinically easy to use and involves standard assessments, used routinely by medical and nursing staff while managing acute asthma.

In order to detect a difference between the two groups at 60 minutes post treatment of 0.5 points on the ASS at the 5% significance level with 80% power, 500 children were required. A difference of 0.5 in ASS was deemed to be the minimum worthwhile clinically important difference to be detected by the research team. There are no studies demonstrating what is a clinically relevant change in an ASS to the patient. As a group of experienced clinicians and researchers we felt a change of 0.5 would be an important difference. There is no evidence base to underpin this pragmatic decision and one of the future studies generated from this work would be to examine what is a clinically relevant change to the patient. Thus the main issue is the clinical relevance of a statistically significant difference in an ASS – this question remains a challenge to those working in acute asthma research.

Severity of asthma exacerbation in the children recruited

We used the BTS definition of ‘severe’ acute asthma and our initial concern was that we were recruiting children into the study who may only have been included because of their tachycardia, especially the younger children. This is an aspect of this definition identified previously as a concern needing further exploration. 64 We did not have comprehensive screening data of the population presenting at the recruitment centres and so external validity of our population could be a concern.

However, data from a national audit of UK asthma admissions of 9428 children, by Davies et al. ,64 between 1998 and 2005, and a recent update of this national audit from November 2011 (J Paton, Royal Hospital for Sick Children, Glasgow, March 2012, personal communication), suggest that we have identified a more severe group of patients. Although the presenting arterial oxygen saturation in air was 94% (IQR 91–96%) in the national audit64 and in this population was slightly lower at 93.6% (range 81–100%), the use of intravenous bronchodilators as a marker of severity, was 4–5% in the national audit64 and in the same level in November 2011 and in this population was 11% (see Table 10). So we feel that we have identified a group of children with acute severe asthma, which does represent the more severe end of the asthma exacerbation population presenting to unscheduled care facilities in the UK and thus our study has external validity.

We now have a data set of over 500 acute episodes of asthma, which will allow us to explore the BTS definitions of severity further, as has been suggested by Davies et al. 64 The magnesium effect is most marked in those children with a more severe exacerbation, as defined here by oxygen saturation in air at presentation. With further analysis of our data using a receiver operating characteristic (ROC) curve we may be able to define where the cut-off point in oxygen saturation at presentation may be to gain maximal effect from the addition of magnesium.

Treatment–covariate interactions

In the initial SAP (see Appendix 3), the plan was to formally test a treatment–covariate interaction for the effect of age by including the interaction term in a regression model. Exploratory analysis was planned to examine the impact on any treatment effect of other factors, such as gender or presenting clinical signs. However, blinded to the results, the treatment–covariate interaction hypotheses were discussed further by the statistical and clinical leads (PW, RD, CP, ID) and several changes to the SAP were made as we felt that this approach would be more robust (see Appendix 3, Changes to statistical analysis plan). Treatment–covariate interactions were investigated for two clinically important baseline covariates, SaO₂ level at presentation and duration of symptoms of the asthma attack. Other factors, such as age or gender, may affect the response but a number of possible patterns of interaction could be argued. Prognostic factors affecting response will be examined in further analysis outside the scope of this report.

Longitudinal assessment

We also assessed the effects of the addition of magnesium to changes in the ASS over 240 minutes. So, rather than a cross-sectional measurement at T60 we were able to see the effect, longitudinally up to 4 hours after treatment. This is a novel approach to assessing ASS and has not been presented in the acute asthma literature before. 39 Longitudinal ASS data were summarised by the AUC. The AUC is a summary measure that integrates repeated assessments over the duration of the treatment.

Figure 3 illustrates the mean longitudinal profiles over the first hour. There was no significant difference in AUC over the first hour during the treatment regimen (p = 0.21; see Table 9). However when we examined the effect over 240 minutes, even accounting for missing values and dropouts we can demonstrate that the statistically significant effect seen at the cross sectional T60 measurement (see Table 8) is sustained up to 240 minutes (see Appendix 5, Figures 15 and 16, and Table 46). Again, the clinical significance of this difference is unlikely to be important [treatment effect on ASS 0.20 (95% CI 0.01 to 0.40)], but it does emphasise that there is a pharmacological effect and that this is sustained over 240 minutes in the overall group. This effect would need to be explored further to examine the magnitude and length of the effect in those with a more severe attack and shorter duration of symptoms. The data from MAGNETIC will allow further exploration of the AUC as a potential core outcome for future acute asthma studies.

Secondary outcomes

We examined secondary outcomes frequently measured in acute asthma studies32,39 (see Appendix 1): need for intravenous bronchodilator therapy, need for PICU admission and intubation, stepping down of treatment after 1 hour, length of stay and additional bronchodilators given. We found no evidence of a difference between those who received magnesium and those who received standard therapy. No paediatric studies of nebulised magnesium have found any evidence of differences in these outcomes but none, including the current study, are powered individually to do so.

The only ‘new’ outcome reported in this study is the ‘stepping down’ of treatment from nebuliser to spacer. We were unable to detect a difference between the two groups: 33% in the magnesium group and 30% in the placebo group (p = 0.53). In the study by Kelly et al. ,43 among 720 patients (adults and children) presenting to 36 EDs in Australia, asthma severity improved from severe to moderate after an hour of treatment in 50%, resulting in a change from nebuliser to spacers – thus ‘step-down’. Stepping down is thus considered to be a proxy for the clinician considering the child to be clinically better and is based, rather than on a score, on a clinician's subjective impression. However, the fact that only one-third stepped down at 1 hour in this study would suggest that we have a group of children with more severe acute asthma attacks than the mixed population of all levels of severity in those presenting to EDs in the Kelly et al. study;43 the mixed age groups and wider spectrum of severity may explain the difference. This concept of stepping down of treatment needs to be further developed for further studies in acute asthma.

An outcome that we did not analyse is the concept of mean duration in supplemental oxygen. Khashabi 200837 (presented in abstract form only) examined 40 children with acute asthma (mean age 3.55 years) but found no difference in an ASS 1 hour after two doses of either nebulised magnesium and salbutamol or salbutamol and placebo, but they did find a difference in mean duration of supplemental oxygen therapy (not defined in the abstract): 15.2 hours (95% CI 9.3 to 21.5 hours) compared with 19.0 hours (95% CI 12.4 to 25.8 hours). 37 This outcome should be defined and explored further in future studies of acute asthma.

Centre effect

A sensitivity analysis was performed to investigate the robustness of ignoring a centre effect in the primary analysis. Two models were fitted when the centre was treated as either a fixed effect or as a random effect. Both models were adjusted for baseline ASS. Reassuringly, there was no evidence that the treatment effect varies by centre (see Appendix 6, Table 47).

Timing of treatment administration

There could be concern that there was significant variation in the timing of the administration of the study medication in the two groups. Reassuringly, there was no clinically significant deviation in the mean prescribed times between the treatment groups on any of the three occasions (see Table 5) and the mean time to administration in both groups was 5.8 (SD 8.3) minutes after randomisation, 23.4 (SD 5.5) minutes after the first dose and 23.3 (SD 6.2) minutes after the second dose, which was as per the protocol. We had previously stated that 15 minutes leeway was clinically acceptable, and Table 6 has shown that only 53/508 instances were considered to be protocol deviations, with 10% in the magnesium group and 12% in the placebo group.

Dose of magnesium given

We used the same dose of isotonic MgSO₄ for all ages on each of the three administrations in the first hour (2.5 ml of 250 mmol/l, tonicity 289 mOsm, 151 mg per dose). This was the dose used by Hughes et al. 16 in their adult study of 52 patients where they demonstrated a significant effect in lung function improvement at 90 minutes post treatment. 16

The ideal dose for children has not yet been clarified and whether the dose needs to be changed with age/weight or whether one standard dose is sufficient, modulated by the child's tidal volume, is yet to be ascertained. There is clearly a dose–response effect in the guinea pig model of magnesium effect and bronchoconstriction59 with guinea pigs with stable tidal volumes but the examination of this issue has not had any exploration in this acute asthma literature.

In the nebulised magnesium studies including children, so far one dose has been used for all ages but these have differed in frequency, formulation and combination with other bronchodilators (see Appendix 1, Table 39). This illustrates how difficult it is to make any comparison and firm conclusion when comparing the literature. 32 This is also a similar research consideration in the adult data.

• Aggarwal et al. 22 (ages 13–60 years, n = 110) 1 ml MgSO₄ (500 mg) three doses 20 minutes apart with β₂-agonist; total magnesium used: 1500 mg (3 × 500 mg).

• Ashtekar et al. 27 and this study, MAGNETIC (ages 2–16 years, n = 508) 2.5 ml of 250 mmol/l, tonicity 289 mOsm, 151 mg per dose; total magnesium used: 453 mg (3 × 151 mg).

• Drobina et al. 23 (ages 12–60 years, n = 110) 125 mg MgSO₄ 0.25 ml of 50% solution (three doses 20 minutes apart with β₂-agonist; total magnesium used: 375 mg (3 × 125 mg).

• Khashabi et al. 37 (ages mean age 3.55 years) two doses of isotonic MgSO₄ not stated.

• Mangat 199814 (ages 12–60 years, n = 33) 3 ml (95 mg) MgSO₄ (four doses, 20 minutes apart) compared with β₂-agonist; total magnesium used: 380 mg (4 × 95 mg).

• Mahajan et al. 18 (ages 5–17 years, n = 62) 2.5 ml isotonic MgSO₄ solution (6.3%) with single dose of β₂-agonist (dose).

• Meral 199619 (ages < 16 years, n = 40) 2 ml of MgSO₄ 280 mmol/l).

No studies have examined the use of frequent doses of nebulised magnesium outside the first hour of treatment. Dose used and frequency given need further research in the clinical setting of acute asthma in children.

Different nebulisers and outputs

This was a pragmatic study and thus did not define a standard nebuliser for each centre but they were all oxygen driven from wall oxygen supplies. We felt that in order to produce a generalisable result we should use what is currently being used in the EDs and CAUs in the UK. Each centre used the same type of nebuliser for all patients within that centre, but different types of nebuliser were used in different centres. There are some American data to suggest that there is variable output from different nebulisers. 69 The PARI LC Star^® (PARI Respiratory Equipment, Midlothian, VA, USA) had an appropriate particle size distribution but a very slow aerosol output rate. The Omron MicroAir^® [Clement Clarke International (CCI), Harlow, UK] had an even slower output rate and a larger particle size distribution, which would be inappropriate for smaller children. In vitro lung deposition with the Aeroneb Go with Idehaler (Aerogen, Galway, Ireland) was 16.0 ± 0.4 mg/minute in older children and approximately one-fifth of that in toddlers. This presumably relates to lung deposition and not necessarily therapeutic effect; some effect may be due to absorption across mucous membranes and independent of lung deposition. Their conclusion was that the Aeroneb Go with Idehaler was the ideal one for a nebulised magnesium study currently under way in the USA.

Unblinding of randomised treatments during the study and protocol deviations and missing values

Interpretation

There are sufficient data in this study to support the use of nebulised isotonic MgSO₄ at the dose of 151 mg given three times in the first hour of treatment as an adjuvant to standard treatment, when a child presents with an acute episode of severe asthma. The response is likely to be more marked in those children with more severe attacks and with a shorter duration of exacerbation. Although the study was not powered to demonstrate this, the data certainly support the hypotheses that nebulised magnesium has a greater clinical effect in children who have more severe exacerbation with shorter duration of symptoms.

Implications for health care

The results of the base-case economic analyses suggest that, from an NHS and Personal Social Services perspective, the addition of magnesium to standard treatment may be cost-effective compared with standard treatment only, though there remains substantial uncertainty around this finding. The results of both sets of analyses (CEA and CUA) show that the probability of magnesium being cost-effective is over 60% at cost-effectiveness thresholds of £1000 per unit decrement in ASS and £20,000 per QALY gained respectively; it is noted that for some parameter variations this probability is much lower, reflecting the impact of variation on the small differences in QALY and costs seen in this trial.

Recommendations for research

Further studies on dose–response at different ages and frequency of administration during an attack are required. The effect on secondary outcomes such as need for intravenous bronchodilators and PICU admissions and length of stay with different nebulised magnesium treatment regimen (dose and frequency) needs further exploration. The concept of different phenotypes and severity where the use of nebulised magnesium can be tailored to the features of the exacerbation needs further exploration.

Currently, three further analyses are planned using these data:

1. exploration of the relationship between ASS and the BTS definition of acute severe asthma

2. assessment of the value of the AUC analysis of ASS

3. examination of the concept of acute phenotypes of asthma in children and the response to treatment.

It may be that these data are sufficient to recommend that nebulised magnesium is added to standard treatment, particularly in those who have a severe attack and those with a short history. Further studies of dose–response pharmacokinetics and frequency of doses, nebuliser use, compatibility studies and animal models to clarify the mechanisms of magnesium use are also to be considered.

Setting trial in context of existing research

The results of this large study are timely. One large study in adults, the 3MG study, is coming to a conclusion4 and another paediatric study in the USA is currently under way. 69 There are limited trial data in children, with only four published studies14,18,19,27 (including the pilot study MAGNET27). This is the largest study of nebulised MgSO₄ in children to date. These data will add further evidence, which may help to improve and strengthen the recommendations of national and international guidelines on the management of acute asthma in childhood.

Study design

This is a multicentre, randomised, placebo-controlled study involving 20–25 sites throughout the UK that plans to recruit 500 children, 250 into each of the study arms. All patients recruited into the study will have standard treatment as per BTS guidelines plus either nebulised MgSO₄ (2.5 ml of isotonic nebulised MgSO₄) or placebo (2.5 ml of isotonic nebulised saline). Each site randomises patients to one of two treatment arms in a 1 : 1 ratio.

Study objectives

The main objective is to compare the ASS at 1 hour of children with acute severe asthma given nebulised MgSO₄ when used as an adjunct to nebulised salbutamol and ipratropium bromide to those given nebulised salbutamol, ipratropium bromide and placebo. The proportion of patients who required a ‘stepping up’ of medication at 1 hour, progression to intravenous treatment, intubation and/or admittance to HDU/PICU will be compared between the two groups.

Secondary objectives are:

Does nebulised MgSO₄ used as an adjunct to nebulised salbutamol and ipratropium bromide for 1 hour in children with acute severe asthma, when compared with nebulised salbutamol, ipratropium bromide and placebo, have an effect on:

1. (a) clinical outcomes in terms of additional treatment/management while in hospital

2. (b) length of stay in hospital

3. (c) patient outcomes in terms of quality of life, time off school and health-care resource usage over the following month

4. (d) parent outcomes in terms of time off work over the following month

5. (e) overall cost to the NHS and society.

Primary outcome

Asthma Severity Score after 60 minutes of treatment.

Secondary outcomes

Clinical (during hospitalisation):

• ‘stepping down’ of treatment at 1 hour, i.e. changed to having hourly treatment after the initial three 20-minute nebulisers

• number and frequency of additional salbutamol administrations

• length of stay in hospital

• requirement for intravenous bronchodilator treatment

• intubation and/or admission to a PICU.

Patient outcomes at follow-up (1 month):

• paediatric quality of life (PedsQL^™) asthma module parental report for all children and self-completion if aged > 5 years, EQ-5D

• time off school/nursery

• health-care resource usage (e.g. GP visits, additional prescribing).

Parent outcomes at follow-up (1 month):

• time off work (related to child's illness).

Inclusion criteria

Severe acute asthma as defined by the BTS/SIGN guidelines. 3

For children ≥ 6 years, severe asthma is based on at least one of the following criteria being met:

1. (a) oxygen saturations of < 92% while breathing room air

2. (b) too breathless to talk

3. (c) heart rate greater than 120 b.p.m.

4. (d) respiratory rate of > 30 breaths per minute

5. (e) use of accessory neck muscles.

For children aged 2–5 years, severe asthma is based on at least one of the following criteria being met:

1. (a) oxygen saturations of < 92% while breathing room air

2. (b) too breathless to talk

3. (c) heart rate greater than 130 b.p.m.

4. (d) respiratory rate > 50 breaths per minute

5. (e) use of accessory neck muscles.

Exclusion criteria

1. (a) coexisting respiratory disease such as cystic fibrosis or chronic lung disease of prematurity

2. (b) severe renal disease

3. (c) severe liver disease

4. (d) known to be pregnant

5. (e) known to have had a reaction to magnesium previously

6. (f) parents who are unable to give informed consent

7. (g) previously randomised into MAGNETIC trial

8. (h) patients who present with life-threatening symptoms

9. (i) previously or currently involved with a trial of a medicinal product in the 3 months preceding screening.

Sample size

In order to detect a difference between the two groups at 60 minutes post treatment of 0.5 points on the ASS at a 5% significance level with 80% power, 500 children are required. This assumes an SD of 1.95, based on a similar population in Australia. 30 The SD was estimated from the Cardiff pilot study27 (EudraCT number: 2004–003825–29) to be 1.7. The target of 500 children will stand. ASS can range from 0 to 9. A difference of 0.5 is deemed to be the minimum worthwhile clinically important difference to be detected. It is a relatively small difference given the low cost and perceived good safety profile of the intervention.

Recruitment

The date the first patient recruited was 3 January 2009. Expected date of end of recruitment and expected date of end of follow-up will be 31 October 2010 and 31 December 2010, respectively. There are 30 sites recruiting patients into the trial and the proposed recruitment targets are given in Table 1.

Representativeness of study sample and patient throughput

Details of patients assessed for eligibility, those who meet the study inclusion criteria, those who are eligible and randomised, those who are eligible but not randomised (with reasons as far as possible), those who withdraw from the study after randomisation (with reasons as far as possible) and those who are lost to follow-up (with reasons as far as possible) will be summarised in a CONSORT flow diagram. Eligible patients who are randomised will be described with respect to demographic details and history (gender, age at randomisation, age at asthma onset, current asthma medication, allergy history, previous admission for asthma, duration of the most recent asthma attack, treatment/nebulisers received pre-admission and ASS, SaO₂, blood pressure, respiratory rate, oxygen therapy at baseline). The number of ineligible patients randomised will be reported.

Baseline comparability of randomised groups

Patients in each treatment group (magnesium and placebo) will be described separately with respect to gender, age at randomisation, age at asthma onset, current asthma medication, allergy history, previous admission for asthma, duration of the most recent asthma attack, treatment/nebulisers/steroids received pre-admission and ASS, SaO₂, blood pressure, respiratory rate, oxygen therapy at baseline. Tests of statistical significance will not be undertaken for baseline characteristics; rather the clinical importance of any imbalance will be noted.

Follow-up assessments and losses to follow-up

The number (and percentage) of patients with scheduled follow-up assessments at 20, 40, 60, 120, 180 and 240 minutes post randomisation will be reported by treatment group. The number lost to follow-up within each treatment group will be reported and reasons where known will be documented in the CONSORT flow diagram. Any deaths and their causes will be reported. Any unblinded events will be reported. The rate of patient and parent outcome questionnaires return at one month will be reported by treatment group.

Description of compliance with therapy

In this study, treatment should be directly observed. Deviations from intended treatment (e.g. withdrawals from randomised treatment) will be summarised for each treatment group. The distribution of timing of treatment administration will be summarised by treatment groups.

Internal pilot

The SD that was used for the original sample size calculation will be checked after approximately 30 patients have been randomised.

The only outcome data that will be analysed within the interim analyses will be the primary outcome of the study which is defined in the protocol as the ASS after 60 minutes of treatment.

This blinded internal pilot will not have any significant impact on the final analysis. 76

Interim analysis plan

In order to estimate the effect of nebulised MgSO₄ for the primary efficacy outcome at each interim and final analysis, the Haybittle–Peto approach will be employed for one interim analysis, planned after approximately 250 children have been randomised, with 99.9% CIs calculated for the effect estimate. This method has been chosen to ensure that interim efficacy results would have to be extreme before early termination is recommended in order to be convincing to the clinical community. The method also minimises controversy regarding interpretation of the results from estimation and hypothesis testing at the final analysis. No inflation factor needs to be applied to the sample size using this approach.

If the trial is stopped early then the analysis will contain all the patients that have been randomised up until that point. The procedures that are described in the statistical quality assurance standard operating procedure will all be implemented before and after the interim analyses.

Intention-to-treat analysis of efficacy outcomes

To provide a pragmatic comparison of the policies of the different drug treatments, the principle of invention to treat, as far as is practically possible, will be the main strategy of analysis adopted for the primary end point. These analyses will be conducted on all patients who have primary outcome data, assigned to the two treatment groups – magnesium or placebo as randomised – regardless of the study treatment or non-study treatment received. A sensitivity analysis will be applied for any missing primary outcome data (see Data analysis, Analysis of missing primary outcome data, below).

Analysis of safety outcomes

For the analysis of safety outcomes, all patients who have received at least one dose of the study drug and were available for follow-up will be included. Patients will be included in the treatment group they actually received.

Analysis of primary efficacy outcome

The primary endpoint is the ASS at T60.

The primary analysis will follow the ITT approach. The hypothesis of no difference between the two treatment arms will be tested using ANCOVA. A p-value of 0.05 (5% level) will be used to declare statistical significance and 95% CIs of the estimated effects will be reported. The primary analysis using ANCOVA will not adjust for any missing data. However, reasons for missing outcome data will be reported and a sensitivity analysis will be undertaken (see Data analysis, Analysis of missing primary outcome data, below).

The assumptions that are made when using ANCOVA (i.e. normality of ASS at treatment levels, homogeneity of variance, homogeneity of regression slopes, linear regression) will be assessed. Histogram of ASS will be plotted for checking normality and a suitable transformation (e.g. square root, log) will be considered to correct non-normally distributed data. Levene's test will be used to test the assumption of homogeneity of variance. Assumptions of linear regression (magnitude of the scatter of the points is the same throughout the length of regression line) and homogeneity of regression slopes (direction and strength of this relationship must be similar in each treatment group) will be detected by examining simple scatterplots between ASS and covariates. If unequal variances, non-linearity and/or non-parallel slopes are present, a suitable transformation of ASS will be used to improve the linearity and to promote equality of the variances.

Randomisation is stratified by centre; however, owing to the large number of small centres, centre will not be included in the model as a covariate, and this is due to the fact that including a large number of small centres may lead to unreliable estimates of the treatment effect and p-values that may be too large or too small. 77 To test the robustness of ignoring the centre effect in the primary analysis, sensitivity analyses will be performed. A GLM type II analysis will be carried out with treatment, centre and treatment-by-centre interaction and baseline measurement included as covariates. Centre will be treated as both fixed and random in separate analyses to assess if there is any effect of this assumption. If the sensitivity analysis suggests the results are not robust to how centre is handled in analysis, centre characteristics (e.g. university hospital, DHS, specialist centre) will be explored further.

All longitudinal ASS data collected will be used in a secondary analysis, with a resulting increase in power. Longitudinal ASS data will be summarised by the AUC. The AUC is a summary measure that integrates repeated assessments of a patient's end point over the duration of the treatment. AUC measures preserved discriminant validity in treatment comparisons and reported more precise treatment effect estimates. 78,79 As the study drug is aimed to lower the ASS over three time intervals, AUC is the most appropriate measure for the treatment comparison.

Analysis of secondary efficacy clinical outcomes

The five clinical secondary outcomes of interest are:

• ‘stepping down’ of treatment at 1 hour

• number and frequency of additional salbutamol administrations

• length of stay in hospital

• requirement for intravenous bronchodilator treatment

• intubation and/or admission to a PICU.

The proportion of patients who required a ‘stepping up’ of medication at 1 hour, progression to intravenous treatment, intubation and/or admittance to HDU/PICU will be compared between the two arms using a chi-squared test. As these are centre-specific outcomes, a sensitivity analysis will be undertaken to account for centre characteristics.

The mean (SD) or median (IQR) of number (frequency) of additional salbutamol administrations will be computed depending on whether it is skewed or not, and compared across treatment groups using a t-test or Mann–Whitney U-test.

Summaries of length of stay in hospital will be presented as means (SDs) or medians (IQRs) depending on whether it is normally distributed or not, and compared across treatment groups.

A formal test of a treatment–covariate interaction will be conducted for the effect of age (2–5 years and ≥ 6 years) by including the interaction term in a regression model. Exploratory analysis will be conducted as to the impact on any treatment effect of other factors such as gender or presenting clinical signs.

A p-value of 0.05 (5% level) will be used to declare statistical significance and 95% CIs of the estimated effects will be reported.

Handling missing health economic data

The ICE command within Stata (version 10.0) will be used to impute missing data for economic outcomes. Following the methods of Briggs et al. 56 for handling missing data, five imputed data sets will be generated through multiple imputation using non-parametric bootstrapping80 in Microsoft Excel 2003 (Microsoft Corporation, Redmond, WA, USA) and the results will be combined using equations described by Briggs et al. 56 to calculate SEs around mean costs and effects that incorporate uncertainty around imputed values as well as sampling variation. SEs will be used to calculate 95% CIs around total and incremental costs and QALYs based on Student's t-distribution.

Cost-effectiveness acceptability curves (CEACs)57 showing the probability that nebulised MgSO₄ is cost-effective relative to placebo at a range of ceiling ratios will be generated based on the proportion of bootstrap replicates (across all five imputed data sets) with positive incremental net benefits. 58 Incremental net benefit can be defined as the incremental QALY gain multiplied by the ceiling ratio minus the incremental cost58 where the ceiling ratio (or threshold) represents the maximum society is willing or able to pay for each additional QALY. All statements about cost-effectiveness will be based on a £20,000 per QALY gained threshold. The probability of nebulised MgSO₄ being less costly or more effective will be based on the proportion of bootstrap replicates that have negative incremental costs or positive incremental benefits, respectively. No discounting will be applied to costs and health effects as the time horizon for the economic evaluation will be < 1 year.

A series of multiway and probabilistic sensitivity analyses will be performed to explore the implications of uncertainty surrounding variables with a degree of uncertainty.

Analysis of missing primary outcome data

Three nebulised study treatments will be given at T0, T20 and T40. The primary analysis will be of the ASS at T60. To investigate how sensitive the results of the primary analysis are to missing data a number of strategies will be used. These sensitivity analyses will involve joint modelling as well as imputing values for missing ASS at T60.

These sensitivity analyses will be carried out as secondary analyses of the study data. The results of these analyses will be compared with the relative effect of missing data on the conclusions of the primary analysis.

Safety analysis

All AEs and SAEs reported by the clinical investigator will be presented, identified by treatment group. AEs will be grouped according to a pre-specified AE coding system and tabulated. The number (and percentage) of patients experiencing each AE/SAE will be presented for each treatment arm categorised by severity. For each patient, only the maximum severity experienced of each type of AE will be displayed. The number (and percentage) of occurrences of each AE/SAE will also be presented for each treatment arm. No formal statistical testing will be undertaken.

Section 7.2: One change

(1) Treatment–covariate interactions

Treatment–covariate interactions were investigated for two clinically important baseline covariates, duration of the most recent asthma attack and SaO₂, owing to reasons explained above (see Chapter 3, Assessing the evidence for treatment–covariate interactions, in the report). It was originally planned to conduct a formal test of a treatment–covariate interaction for the effect of age. Although age may affect the response, a number of possible interactions could be argued.

Section 9: One change

(1) Timing of treatment regimes

Protocol deviation was originally defined as deviations outside acceptable timing window (T = 60 + 15 minutes) without explanation. However, because the prescription time of each treatment was reported rather than the time of the end of the third treatment, it was only possible to determine the difference in prescription times between the first and third treatment which should be ≤ 55 (40 + 15) minutes. Therefore, if this timing was > 55 minutes, this was defined as a deviation outside the acceptable window.

Amendments from version 6.0 (23 July 2009) to version 6.1 (18 January 2010)

Amendments from version 5.0 (19 September 2008) to version 6.0 (23 July 2009)

Amendments from version 4.0 (18 April 2008) to version 5.0 (19 September 2008)

Amendments from version 3.0 (3 March 2008) to version 4.0 (18 April 2008)

Amendments from version 2.0 (18 January 2008) to version 3.0 (03 March 2008)

Amendments from version 1.0 (23 November 2007) to version 2.0 (18 January 2008)